DOES EXPORTING MATTER FOR POVERTY REDUCTION: THE CASE OF KENYAN MANUFACTURING NANCY NELIMA NAFULA MWANGE Kenya Institute for Public Policy Research and Analysis, Nairobi, P.O Box 20107 00100 Kenya (e-mails: [email protected], [email protected], [email protected]JEL Classification numbers: F14, I32, L60 Key words: Poverty, inequality, exports April 2013
49
Embed
DOES EXPORTING MATTER FOR POVERTY REDUCTION: THE …
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
DOESEXPORTINGMATTERFORPOVERTYREDUCTION:THE
CASEOFKENYANMANUFACTURING
NANCYNELIMANAFULAMWANGE
Kenya Institute for Public Policy Research and Analysis, Nairobi, P.O Box 20107 00100
LIST OF TABLES............................................................................................................. iii
ABBREVIATIONS AND ACRONYMS ......................................................................... iv 1. Introduction .............................................................................................................. 6
2. Poverty and inequality .............................................................................................. 7
3. Theoretical and empirical issues ............................................................................ 12
1.3.1 Poverty: identification, aggregation and measurement ......................12
1.3.2 Foster Greer and Thorbecke (FGT) poverty measures ......................14
0 and 1 are coefficients and 32 , and 4 are vector coefficients to be estimated
≥0 = poverty aversion parameter.
But we know that exports are potentially endogenous in the poverty equations (7a) and
(7b). We therefore specify export equations (8a) and (8b) respectively.
Suppose in the reduced form equation for export propensity, we have:
111 ZY ………………………………………………………………………….. (8a)
1Y is export propensity
Z is a set of exogenous variables including exclusion restrictions
1 is error term
We specify a similar functional form with export intensity see equation (7b).
The reduced form equation for export intensity is given by:
222 RY ………………………………………………………………………… (8b)
Where:
2Y is export intensity
R is a set of exogenous variables including exclusion restrictions
2 is error term
5. Definition of variables
1.5.1 Dependent variables
We define the dependent variables using the indices derived by Foster, et al., (1984),
namely, the poverty incidence, poverty gap and poverty severity.
20
Poverty incidence (also prevalence of poverty or the poverty headcount ratio) is measured
as the proportion of individual workers below the poverty line, i.e., the poor expressed as a
proportion of the population (number of workers). Poverty incidence is measured as a
dummy variable, which takes the value “1”if the individual is poor and a value of “0” if
the individual is non-poor.
Poverty gap (also average income shortfall/poverty depth) is measured as the proportional
shortfall of the average poor person from the poverty line. Poverty gap is a proportion.
Poverty gap squared (poverty severity) is measured by squaring the poverty gap see
Foster, et al., (1984) and Mwabu, et al., (2003) for further details on FGT poverty
measures. The poverty gap and poverty gap squared only apply if an individual is poor.
This means we have to control for selection bias when estimating these two poverty
equations.
1.5.2 Explanatory variables
The key independent variables are measures of probability of exporting and proportion of
exports in the total output. The first independent variable, export propensity, is defined as
the probability that a firm enters a foreign market. Export propensity is measured as a
dummy variable which takes the value “1” if a firm is engaged in exporting and a value of
“0” if the firm does not export (only sells in the domestic market). The second
independent variable, export intensity, is defined as the proportion of the total sales value
of the firm that is exported. This is measured by dividing the total value from exports by
the total value of the firm’s output.
Other explanatory variables consist of a number of worker demographic characteristics
that could affect poverty such as age, years of schooling, and also a vector of firm level
characteristics that impact on poverty, such as size (number of employees), and location
dummies.
21
6. Data and summary statistics
We use firm-level data from the 2002/3 Regional Program on Enterprise Development
(RPED) Survey to analyse the effects of mental health (employee concerns about
HIV/AIDS) on performance of firms and on wages of the industrial workers in Kenya.
The Kenyan manufacturing sector is classified under three main sub-sectors, namely,
agro-based, engineering and chemical and mineral clusters. The agro-based sub-sector has
developed on the basis of domestic resource activities and contributes 68 percent of the
manufacturing sector value added. The engineering sub-sector relies heavily on imported
raw materials and contributes about 12 percent of the manufacturing sector value-added.
The chemical and mineral sub-sector is Research and Development oriented and
contributes 20 percent of the manufacturing sector value-added.
While firm-level data sets are well established for most of the Organisation for Economic
Co-operation and Development countries, corresponding data of good quality are hardly
available for most developing countries, Kenya included. Considerable advances have
been made by the World Bank with the ‘Regional Program on Enterprise Development
(RPED) Surveys in making firm level data available in developing countries. The RPED
offers harmonized cross-sectional data on the investment climate, i.e., the conditions
affecting firm production and investment behaviour, in developing countries.
Firm level panel data would be better suited for this study since problems of endogeneity
resulting from explanatory variables that are firm-specific and possibly correlated with
mental health capital, could be tackled by using appropriate time lag structures.
Unfortunately, the existing RPED panel data sets (1993-1995) available for most sub-
Saharan African countries including Kenya do not have health data, key information for
this analysis.
The Kenyan 2002/2003 RPED dataset is therefore an interesting alternative source of
health data for this study, despite its limitations in other dimensions. The Kenyan RPED
22
was organized and coordinated by the World Bank. It was executed in 2003 by Kenya
Institute for Public Policy Analysis (KIPPRA) in collaboration with the Kenya National
Bureau of Statistics (KNBS). The RPED 2002/03 survey of 282 formal manufacturing
firms and workers covered seven sub-sectors in five urban areas, namely, Nairobi,
Mombasa, Eldoret, Kisumu and Nakuru.
The information on individuals was obtained through interviews, with at most ten
employees randomly chosen from a list of workers of each firm. A study by Mairesse and
Greenan (1999) shows that, econometric studies of the firm can be effectively and
substantially enriched by using information collected from employees, even if only a few
of them (at least two) are surveyed per firm. Though variables measured on the basis of
the answers of very few employees per firm are subject to very important sampling errors,
they can be usefully included in a measurement model implemented with firm level data
(Addison and Belfield, 2004, Bigsten et al., 2000, Corvers, 1997, Soderbom and Teal,
2001).
The information on the firms was elicited from representatives of each firm. The data set
on which this research is based does not contain information on individual HIV status or
on deaths due to AIDS. None the less, the respondents are well aware of the epidemic.
Majority are familiar with the symptoms of AIDS, are aware of how HIV/AIDS is
transmitted, know where to go for HIV tests, know their own behaviour and may be their
spouse behaviour to understand whether they are at risk of HIV infection or not.
Some of the information collected include: ownership structure, total sales revenue, value
of exports, total number of employees, absenteeism, proportion of employees believed to
be HIV positive, and proportion of employees believed to have died of HIV/AIDS. The
employees interviewed provided a range of information including education level,
previous experience, experience in the current firm, age, sex, hours of work, degree of
concern about HIV/AIDS, willingness to test for HIV/AIDS, job tenure length, own-
financed training, firm supported on–the–job training, previous training before joining the
23
firm, health status, days missed work due to own illness, days missed work due to family
or friend’s illness, wages received, benefits received, and numerous other personal
characteristics.
Descriptive statistics are shown in Table 2.
Table 2: Summary statistics: essay three Variable Observations Mean Standard
deviation Age of workers (in years) 1863 36.03 9.62 No schooling 1825 0.01 0.11 Primary 1825 0.19 0.40 Secondary 1825 0.43 0.49 Technical and vocational 1825 0.26 0.44 University 1825 0.10 0.31 Years of education of workers 1821 11.34 3.89 Monthly wage per worker 1863 17243 29715.39 Dummy for export (export propensity) (1=yes) 1918 0.54 0.50 Poverty headcount ratio ( 0P ) (1=yes) 1863 0.037 0.19
Poverty depth ( 1P ) 1863 0.01 0.09
Poverty Severity ( 2P ) 1863 0.01 0.07
Firm size (total number of employees) 1863 201.44 324.85 Predicted probability of exporting 1821 0.54 0.16 Log of investment last year 1863 8.32 7.65 Inverse of Mills ratio 1821 2.29 0.33
Source: RPED survey 2002/3.
The data shows that very few workers have no formal education, those with primary
education are about 19.5 percent. Majority of the workers have secondary education (43
percent) while technical education and university account for 26.2 percent and 10.4
percent respectively. On average, workers earn Ksh. 17,243 and about 54 percent of firms
are involved in exporting. The high mean wage may suggest that there is no wage poverty
in manufacturing firms. However, this is not true, poverty statistics show that about 4
percent of workers earn a monthly wage below the poverty line. This may imply high
income inequalities among the workers. Most firms are medium sized with about 201
workers.
24
A further analysis of correlation shows that all the poverty indicators are negatively
correlated with age, years of worker education, firm-size and last year investment, while
exports are positively correlated with the same variables. Worker years of education and
log of previous investment by the firm are both significant at 1 percent. While firm size
correlated with exports is significant at 1 percent, firm-size correlated with the three
poverty indicators is significant at 10 percent. Age is not significant across all the cases.
Table 3: Pair-wise correlations between exports, poverty and selected characteristics of workers and firms (p-values in parentheses) Variables Exports P0 P1 P2 Age of the worker (in years) 0.0208
(0.3698) -0.0307 (0.1859)
-0.0292 (0.2072)
-0.0292 (0.2072)
Worker years of education 0.1183 (0.0000)
-0.1158 (0.0000)
-0.1157 (0.0000)
-0.1157 (0.0000)
Log of Firm size (total number of employees)
0.3179 (0.0000)
-0.0388 (0.0943)
-0.0387 (0.0952)
-0.0387 (0.0952)
Log of investment last year 0.2190 (0.0000)
-0.0809 (0.0005)
-0.0813 (0.0004)
-0.0813 (0.0004)
Source: RPED survey 2002/3.
The correlations in Table 3 show that the association between years of education and
poverty indicators and export propensity is highly significant at 1 percent level. For
instance, a 10 percent increase in the years of education is associated with a 1.2 percent
increase in exports. Similarly, a 10 percent increase in the log of lagged investments is
associated with 2.2 percent increase in exports. These correlations are symmetric.
7. Poverty and inequality profiles
Poverty and inequality is determined to a large extent by characteristics that define the
endowments and potentials of individuals, households or communities/firms. Differences
in the human and physical capital of workers affect the pattern of wage income in the firm.
In a firm environment, these characteristics include export status of the firm, location of
firm, gender of worker and highest education level attained. Table 4 shows the poverty
profile of manufacturing workers. Even without sharing their incomes with the other
25
members of household, some workers are still trapped in poverty.
Log of investment last year .0093 (5.94) .0098 (5.85) Age of the worker .0006 (0.54) .0006 (0.50) Log of Worker years of education .0066 (2.12) .0070 (2.08) Log of Firm size (total number of employees) .0865 (8.60) .0902 (8.04) Constant -.0201 (0.29) - F statistics [p-value] 43.79 (0.0000) Adjusted R-squared .0988 Wald chi2(4) 135.79 (0.0000) Pseudo R-squared .0736 No. of observations 1751 1751
Source: RPED survey 2002/03. Note: Absolute t statistics in parentheses. Critical t-values: 1%=2.58, 5%=1.96 and
10%=1.65.
Our main variable of interest, which is also our instrument, is the log of investment last
year. The magnitude of the coefficient on this variable is .0093 and .0098 for OLS (LPM)
and the probit model, respectively. The coefficient is positive as expected and the t-
statistic is significant at 1 percent. According to the LPM estimates, a percentage increase
29
in the previous investment would increase the probability of exporting by 0.0093. The
results show that previous investment is a relevant instrument for exports. The diagnostic
tests show a very high magnitude for first stage F-statistic (35.84 [0.0000]) on the
instrument. The F-statistic is greater than ten. This suggests that the instrument is strong
and valid for identification (Godfrey, 1999, Nevo and Rosen, 2010, Shea, 1997, Staiger
and Stock, 1997).
1.8.2 The impact of exporting on poverty status
Table 7 compares the OLS (LPM) without controls for endogeneity and heterogeneity
with results obtained from the control function approach accounting for endogeneity and
heterogeneity. We use the control function approach to control for endogeneity and
unobserved heterogeneity as shown in columns 3, 4 and 5. We checked the endogeneity of
exports using Durbin-Wu-Hausman’s test of endogeneity.
A comparison of the two results from columns 1, 2 and 3 shows that controlling for
unobservables in the estimation of parameters of the export function, makes a difference.
The estimated results follow a prior expectation regarding the bias caused by endogeneity
problem. When this bias is not controlled for, the coefficients associated with export
variables are expected to be biased upward. Thus, controlling for endogeneity biases
seems to be important since differences in the magnitudes of the coefficients arise. The
years of education attained by the worker is an important factor in poverty reduction. The
higher the years of education the less likely it is for a worker to live in poverty. The
coefficient is very significant. Firm size also gives interesting results. The results show
that poverty is prevalent in large firm sizes. This may be interpreted to mean lack of
trickle down effects in such firms. Results from probit function in column 4, give even
more precise estimates: firms which export have poverty rates 26.2 percent lower than
those which do not export, which suggests that exporting is good for poverty reduction.
30
Table 7: The impact of exporting on poverty incidence (dependent variable =poverty headcount)
Variables
Estimation Methods LPM-2SPS without controls for endogeneity and heterogeneity (1)
Control Function Approach LPM (with controls for endogeneity) 2SRI (2)
LPM (with controls for heterogeneity (3)
Probit Marginal Effects (with controls for heterogeneity) (4)
Export propensity - -.1681 (2.70) -.1685 (2.71) -.2617 (2.65) Predicted residual of probability of exporting - .1463 (2.31) .1854 (2.56) .1706 (2.58) Predicted probability of exporting -.1681 (2.50) - - Predicted residual interacted with exporting propensity
- - -.0831 (1.50) -.0534 (1.01)
Age of the worker ( in years) -.0007 (1.36) -.0007 (1.25) -.0007 (1.24) -.0006 (1.30) Worker years of education -.0051 (3.86) -.0051 (3.71) -.0050 (3.64) -.0291 (3.32) Log of firm size (total number of employees) .0154 (2.03) .0154 (2.48) .0147 (2.38) .0147 (2.39) Constant .1433 (5.21) .1433 (4.77) .1634 (4.35) - F statistics [p-value] 8.85 (0.0000) 6.05 (0.0000) 5.08 (0.0000) Adjusted R-squared 0.0199 .0227 .0237 Wald chi2(6) - - - 43.98 (0.0000) Pseudo R-squared - - - .0632 No. of observations 1751
Source: RPED survey 2002/3. Notes: Predicted probability of exporting is derived from export equation (8a). Absolute t statistics in parentheses. Critical t-values:
1%=2.58, 5%=1.96 and 10%=1.65.
31
Table 8 compares results from probit estimation which take into account non-linear effect
of the interaction term (column1) with results from probit estimation under the linearity
assumption of the interaction term (column 2) (see Friedrich, 1982, Norton et al., 2004).
The coefficient of the interaction term in the probit model Error! Reference source not
found. is improperly estimated. Using the inteff commad in stata, after running the probit
model, the results are properly computed. Results from the non-linear probit show that the
effect of the interaction term is quite large (coefficient = -0.488, t = -4.26) as opposed to
probit mean effect under linearity assumption (coefficient = -0.053, t = -1.01). Besides, the
former coefficient is highly statistically significant as opposed to the latter which is
statistically insignificant. This means that while the probit estimates in Table 7 show no
signs of heterogeneity, the re-estimation of interaction effect (assuming non-linearity)
show that there is strong evidence of heterogeneity.
Table 8: Probit estimates of the mean coefficient of the interaction term in Error! Reference source not found. Variables Estimated non-linear
mean effect (1)
Estimated mean effect under linearity assumption (2)
Predicted residual interacted with export propensity
-0.488 -0.053
Standard error of the estimated coefficient
0.155 0.0528
z-statistic / t-statistic -4.258 -1.01 No. of observations 1752 1752
Source: RPED survey 2002/3.
In estimating the impact of exporting on poverty depth, we restricted the sample only to
the poor workers. It was difficult to think of a variable (instrument) that will affect poverty
incidence and not poverty depth. Therefore the potential problems of sample selection are
not addressed due to data limitations. Without the exclusion restrictions in the probit
equation, the results should be interpreted with caution.
We present results from OLS and Instrumental Variable (IV) estimations. This approach is
similar to that used by Moll (1996) and Mwabu and Schultz (2000). The results are shown
32
in Error! Reference source not found.. In column (1) OLS estimates without controls for
endogeneity and unobserved heterogeneity show a positive sign on the coefficient for
exports, but this is insignificant while the IV regression in column 3 show that exports
reduce poverty. A 10 percent increase in the proportion of exporting firms would reduce
poverty gap by 8.3 percent. The coefficient is significant at 5 percent level.
Table 9: The impact of exporting on poverty depth (dependent variable =poverty depth) Variables
Estimation Methods OLS (1)
Two stage least squares (one- step command) First stage regression dependent variable = export (2)
Second stage regression dependent variable = poverty depth (3)
Export propensity .0009 (0.23)
- -.0829 (2.54)
Age of the worker ( in years)
.0002 (0.78)
0.0006 (0.54)
.0003 (0.91)
Worker years of education
-.0019 (3.50)
.0066 (2.14)
-.0013 (2.12)
Log of firm size (total number of employees)
-.0009 (0.81)
.0865 (9.55)
.0075 (2.29)
Log of previous investments
- .009 (5.99)
-
Constant .0307 (2.76)
-.0201 (0.30)
.0296 (2.41)
F statistics [p value] 4.00 (0.0031)
47.84 (0.0000)
4.03 (0.0030)
Adjusted R-squared 0.0087 0.0967 - No. of observations 1751
Source: RPED survey 2002/3. Note: Absolute t statistics in parentheses. Critical t-values: 1%=2.58, 5%=1.96 and
10%=1.65.
Similarly, on squaring the poverty gap, the OLS estimates bear a positive coefficient while
the IV estimates show a negative effect of exporting on poverty. The results presented in
Table 10 show that a 10 percent increase in exporting reduces the severity of poverty by
6.2 percent.
33
Table 10: The impact of exporting on poverty severity (dependent variable =poverty severity) Explanatory variables
Estimation methods OLS-estimates (1)
Two stage least squares (one step command) First stage regression dependent variable = export (2)
Second stage regression dependent variable = poverty severity (3)
Export propensity .0042 (1.32)
- -.0622 (2.44)
Age of the worker .0003 (1.45)
.0006 (0.54)
.0004 (1.52)
Worker years of education
-.0010 (2.44)
.0066 (2.14)
-.0005 (1.13)
Log of firm size (total number of employees)
-.0008 (1.26)
.0865 (9.55)
.0058 (2.22)
Log of previous investments
- .009 (5.99)
-
Constant .0085 (1.10)
-.0201 (0.30)
.0076 (0.85)
F statistics [p-value] 2.53 (0.0389)
47.84 (0.0000)
2.26 (0.0604)
Adjusted R-squared/R-squared
0.0075 0.0988 -
No. of observations 1751
Source: RPED survey 2002/3. Note: Absolute t statistics in parentheses. Critical t-values: 1%=2.58, 5%=1.96 and
10%=1.65.
1.8.3 Export intensity and poverty status
In Table 11, we report OLS estimates of the export intensity. The estimation in Column 1
excludes the location dummies while column 2 includes the location dummies. From these
results, we observe that using log of previous investments as an instrument for export
intensity is valid. The quality of our instrument is assessed using tests proposed by Bound
et al.(1995). In this case, the instruments should have a significant effect on the export
34
intensity. Further evidence on the strength and validity of this instrument is tested using
the Shea formula (see Nevo and Rosen, 2010, Shea, 1997, Staiger and Stock, 1997). The
first stage F- statistic (110.04) is incredibly high and statistically significant, suggesting
that previous investment is a strong instrument for export intensity. When estimating the
impact of export intensity on the poverty incidence we maintain log of previous year
investment as the instrument for export intensity. The difficulties of getting a good
instrument are well documented. For instance see Bound et al. (1995).
Estimation method OLS - without location dummies (1)
OLS - with location dummies (2)
Age of the worker ( in years) -.0003(0.44) -.0004 (0.63) Worker years of education -.0065 (4.10) -.0058 (3.86) Log of firm size (total number of employees)
.0770 (13.73)
.0778(15.15)
Log of previous year’s investment .0075(9.61) .0081(10.71) Location dummies No Yes Constant -.1669 (5.07) -.3107(7.74) F statistics [p-value] 74.04(0.0000) 52.47(0.0000) R-squared 0.2098 0.2726 No. of observations 1751
Source: RPED survey 2002/3. Note: Absolute t statistics in parentheses. Critical t-values: 1%=2.58, 5%=1.96 and
10%=1.65.
In Table 12 column 1 we report LPM-2SPS results. The results for LPM-2SPS indicate
that the coefficient for predicted export intensity is negative and significant at 5 percent
level. In column 2, 3 and 4 we estimate the poverty incidence using the control function
approach. We find that the sign on the coefficient for export intensity remains negative.
However, the magnitude declines as we control for endogeneity and heterogeneity in both
LPM and the probit estimations.
35
A percentage increase in the proportion of firms exporting would result in about 19-21
percent reduction in the headcount ratio in manufacturing firms. As shown in earlier
studies, firms that export pay higher wages to their workers as compared to those firms
that do not export, such that other things being constant, exporting reduces poverty.
Table 12: The impact of export intensity on poverty incidence (dependent variable =poverty headcount) Explanatory Variables
Estimation Methods
LPM-2SPS with location dummies (1)
Control Function Approach LPM-2SRI with controls for endogeneity (2)
LPM-2SRI with controls for heterogeneity (3)
Probit Marginal Effects (with controls for heterogeneity) (4)
Source: RPED survey 2002/3. Notes:Residual is derived from equation (8b). Absolute t statistics in parentheses. Critical t-values: 1%=2.58, 5%=1.96 and 10%=1.65.
38
A common concern in the literature is that there is simultaneity bias between poverty and
exports (Balat, et al., 2009). This means that poverty depth affects export intensity and
export intensity affects poverty depth. For this reason, we report additional results that
should be more robust to these potential problems than the OLS estimates. We use the IV
and the control function methods to correct for these biases (Mwabu, 2009, Soderbom et
al., 2006).
The results in Table 14 show that the IV estimates and the control function estimates are
higher in magnitude compared to OLS estimates. The results also suggests that treating
export intensity as endogenous increases its effect on the poverty depth. A test for
endogeneity provides evidence of endogeneity as shown by significant coefficients on
predicted residual for export intensity in column 4, but there is no evidence of
heterogeneity because the coefficient on predicted residual for export intensity interacted
with export intensity in column 5 is statistically insignificant.
An extension of similar analysis of export intensity on the severity of poverty is shown in
Table 15. Poverty severity is the same as poverty depth except that greater weight is
placed on people in extreme poverty. The pattern on export intensification on poverty
severity is similar as previously observed in the estimation of the export intensity on
poverty incidence and poverty depth. The coefficient on the export intensity from OLS
regression has the expected sign and is statistically significant at 5 percent level. However,
the coefficients on export intensity from the IV regression (column 3) and the control
function approach (columns 4 and 5) are substantially larger than those obtained from
OLS regression in column 1. The IV estimates, show it is evident that manufactured
exports are associated with large poverty reduction gains because an increase in mean
export intensity lowers the intensity of poverty by 0.0709.
39
Table 15: The impact of export intensity on poverty severity (dependent variable =poverty severity) Variables
Estimation Methods OLS -without controls for endogeneity and heterogeneity (1)
Two stage least squares (one step command)
Control function approach
First stage regression dependent variable=export intensity (2)
Second stage regression dependent variable=poverty severity (3)
with controls for endogeneity (4)
with controls for endogeneity and heterogeneity (5)
Predicted residual interacted with export intensity
- - - - -.0202 (1.17)
Age of the worker .0003 (1.41) -.0004 (0.63) .0003 (1.35) .0003 (1.38) .0003 (1.35)Worker years of education -.0010 (2.50) -.0058 (3.77) -.0013 (2.84) -.0014 (2.98) -.0015 (3.01)Log of firm size (total number of employees)
Source: RPED survey 2002/3. Note: Absolute t statistics in parentheses. Critical t-values: 1%=2.58, 5%=1.96 and 10%=1.65.
40
The effects of other firm specific variables on the intensity of poverty are also statistically
significant. The coefficient on worker’s years of education from IV regression (column 3)
shows that a 1 percent increase in years of schooling is associated with a 0.00013 percent
reduction in the intensity of poverty. Similarly, the control function results are not so
different from the IV estimates and show different but statistically significant results when
we control for endogeneity and heterogeneity biases.
The coefficient on the predicted residual is significant, suggesting that endogeneity is a
problem. The test for heterogeneity is depicted by inclusion of the interaction variable
(i.e., between the predicted residual of export intensity and export intensity itself). The
coefficient on the interaction term is statistically insignificant, implying that unobserved
heterogeneity is not a problem.
1.8.4 Policy simulations
One of the objectives of modelling export intensity and export propensity of firms is to
simulate the effects of policy interventions that affect exports. The predictive power of the
simulation models depends on the estimated coefficients of the export propensities and
export intensities.
In this section we use results from two sets of tables i.e., Table 12, Table 14 and Table 15
to evaluate and compare the welfare effects of increasing the mean export intensity by 1
percent, i.e., from 0.15 to 0.16. We also use results from Table 7, Table 9 and Table 10 to
evaluate the welfare effect of increasing the mean proportion of firms in the export sector
by 1 percent, i.e., from 0.54 to 0.55. Previous experience has shown that in a span of one
year, about 20 firms (approximately 1 percent) join the export market; thus the simulated
policy induced changes in export propensity is feasible.
Results from this policy analysis are presented in Table 16 together with the resulting
41
changes in poverty incidence, poverty depth and poverty severity. The sample proportions
are the proportions corresponding to each of the poverty indicators computed from the raw
data.
Table 16: Policy simulations Variables Poverty
incidence Poverty depth
Poverty severity
Sample means (percent) 3.7 1.3 0.8
Policy 1: Increase mean export intensity by 1 percent, from 0.15 percent to 0.16 percent Change in poverty level (percent)
-1.9
-0.9
-0.7
Policy 2: Increase proportion of firms that export by 1 percent, from 0.54 percent to 0.55 percent Change in poverty level (percent)
-2.6
-0.8
-0.6
Source: RPED survey 2002/3.
Notes: The decision to base the simulation on 1% rests on the reasonability of the changes in the export variables between 2001 and 2002.
Besides the notable reductions in the poverty headcount ratio, increasing the proportion of
exports in total sales also affects poverty depth and poverty severity. When we vary the
export intensity while holding all the other factors constant, we find that poverty
headcount ratio declines by about 1.9 percent. This means that the proportion of poor
people in the manufacturing sector would decline by 1.9 percent. Given the low poverty
levels in exporting firms, we conclude that exporting substantially reduces poverty in the
manufacturing sector.
Another policy option is to increase the proportion of firms that export by 1 percent. The
overall effect of this policy on poverty is more or less similar to that of increasing the
proportion of exports in total sales. However, the magnitude of the effect on headcount
ratio is slightly higher (2.6 percent). Again, poverty is highly elastic with respect to
exporting. This suggests that poverty in manufacturing firms could be wiped out through
42
policies that boost exports. However, the amount of investment required to bring about
reasonable reduction in poverty could be substantial and may not be easily affordable to
most of the firms, which suggests the need to subsidize firms that have export potential.
In our analysis, we find that results based on proportional changes in poverty indices tend
to exaggerate the effect of exporting on poverty as opposed to results based on level
changes in poverty measures. The reason behind this is that proportional changes tend to
suffer from base effect such that they will vary according to the value of the base figure
and so this is the case in our analysis. Also the common interpretation of poverty changes
is in terms of level changes and not in terms of proportional changes (Foster, et al., 1984).
For this reason we prefer the results for poverty simulations conducted at the level
changes.
We conclude that for a developing country like Kenya, the effect of exports on poverty
matters for export propensity (growth in the export among countries). However, this does
not exclude the poverty reduction impacts of export intensity (growth of exports
conditional on being in an export relationship) only if Africa and in particular, Kenya,
would afford the required amount of investments in exporting. These results are similar to
those of other studies that have analysed the impacts of changes in extensive and intensive
margins of exports (Evenett and Venables, 2002, Hummels and Klenow, 2005).
9. Summary and conclusions
There is limited empirical evidence on the effects of manufactured exports on poverty
levels in Kenya. Previous literature has mainly analysed the effect of various trade policy
instruments on poverty (Haiti National Strategy team, 2006, Soderbom and Teal, 2003).
The purpose of this study is to examine empirical evidence on whether export
intensification could be used as a poverty reduction strategy. The data, drawn from a
survey of Kenyan manufacturing firms, indicate that exporting is associated with low
43
poverty. However, the effect of exporting in reducing the headcount ratio is small
compared to its role in reducing the poverty depth and poverty severity among the poor
people. While this is true, the results also show that there are other factors that are
important in reducing poverty, such as education.
We use the control function approach to remove endogeneity and heterogeneity biases in
the parameter estimates in all the models studied. When biases are not controlled for, the
coefficients on exports are biased upward in a poverty analysis model. Thus, controlling
for biases from endogeneity is important since large differences in the magnitudes of the
coefficients can arise.
The results show that firms which export have poverty rates 26.2 percent lower than those
which do not export. Further analysis shows that a 10 percent increase in the proportion of
exporting firms would reduce poverty gap by 8.3 percent. Similarly, a 10 percent increase
in exporting would reduce the degree of inequality (severity of poverty) by 6.2 percent.
The results for OLS indicate that export intensity is negatively associated with poverty;
the coefficient on export intensity is highly significant. We find similar results when using
control function approach. However, controlling for heterogeneity lowers the size of the
coefficient on export intensity. A percentage increase in the proportion of firm exports
would result in about 19-21 percent reduction in the headcount ratio in the manufacturing
firms.
We use the IV and the control function methods to estimate effects of export intensity on
poverty. We find that IV estimates and control function estimates are larger than the OLS
estimates. Further analysis shows that there is evidence of endogeneity and no evidence of
heterogeneity. The effect of export intensification on poverty severity is similar to the
results obtained for the effect of export intensity on poverty incidence and poverty depth.
The OLS estimates have the expected sign and are significant at 5 percent level. The IV
44
estimates suggest that there are large poverty gains from manufactured exports: an
increase in mean export intensity lowers the intensity of poverty by .0709.
Similarly, the control function results are not so different from the IV estimates but change
significantly when we control for endogeneity and unobserved heterogeneity biases.
While it is important to identify the marginal impact of each of these measures on the
outcome under consideration, it is also important to compare the relative costs of various
policies. The coefficients from the probit and OLS regressions were used in this section to
compute the elasticities of poverty with respect to changes in the policy variables used in
simulations.
Based on our regression results, we simulated effects of two policy interventions to
examine their possible effects on poverty. We simulated the effects of increasing the
proportion of exports in total sales by 1 percent. The impact on poverty was a reduction of
about 1.9 percent for headcount index, 0.9 percent for poverty depth and 0.7 percent for
poverty severity. A further simulation of poverty impact of increasing the propensity to
export by 1 percent showed that the headcount ratio would decline by about 2.6 percent.
Inference on the basis of our regression analysis shows that controlling for other factors, a
rise in exporting significantly reduces the risk of being poor.
We also find that poverty is highly elastic with respect to exporting propensity and
exporting intensity. However, these results should be interpreted with caution since the
computation of elasticity may have been exaggerated due to the base effect. The outcome
is highly dependent on the base value used in computations.
45
REFERENCES
Addison, J. T. and Belfield, C. R. "Unions, Training, and Firm Performance: Evidence from the British Workplace Employee Relations Survey," Discussion Paper No. 1264. Bonn: IZA, 2004.
Ali, A. A. G. and Thorbecke, E. "The State and path of poverty in sub-Saharan Africa." Journal of African Economies, 2000, 9(supplement (1)), pp. 9-40.
Atkinson, A. B. "On the Measurement of Inequality." Journal of Economic Theory, 1970, 2, pp. 244-63.
Balat, J.; Brambilla, I. and Porto, G. "Realising the gains from trade: Export crops, marketing costs, and poverty." Journal of International Economics, 2009, 78, pp. 21-31.
Bigsten, A.; Isaksson, A.; Soderbom, M.; Collier, P.; Zeufack, A.; Dercon, S.; Fafchamps, M.; Gunning, J. W.; Teal, F.; Gauthier, B., et al. "Rates of Return on Physical and Human Capital in Africa's Manufacturing Sector." Economic Development and Cultural Change, 2000, 84(4), pp. 801-27.
Blackorby, C. and Donaldson, D. "Ethical Indices for the Measurement of Poverty." Econometrica, 1980, 48(4), pp. 1053-60.
Bound, J.; Jaeger, D. A. and Baker, R. M. "Problems with Instrumental Variables Estimation When the Correlation Between the Instruments and the Endogenous Explanatory Variable is Weak." Journal of the American Statistical Association, 1995, 90(430), pp. 443-50.
Central Bank of Kenya. "Annual Report." 2006.
____. "Annual Report," 2003.
Chakravarty, S. R. "Ethically Flexible Measures of Poverty." The Canadian Journal of Economics, 1983, 16, pp. 74-85.
Clark, S.; Hemming, R. and Ulph, D. "On Indices for the Measurement of Poverty." The Economic Journal, 1981, 91, pp. 515-26.
46
Collier, P. and Lal, D. "Poverty and Growth in Kenya," World Bank Staff Working Paper No. 389. Washington D.C.: World Bank, 1980.
Corvers, F. "The Impact of Human Capital on Labour productivity in Manufacturing Sectors of the European Union." Applied Economics, 1997, 29, pp. 975-87.
Evenett, S. J. and Venables, A. J. "Export Growth in Developing Countries: Market Entry and Bilateral Trade Flows," University of Bern, 2002.
Foster, J. E.; Greer, J. and Thorbecke, E. "A Class of decomposable Poverty Measures." Econometrica, 1984, 52(3), pp. 761-66.
Foster, J. E. and Shorrocks, A. F. "Subgroup Consistent Poverty Indices." Econometrica, 1991, 59(3), pp. 687-709.
Friedrich, R. J. "In defense of multiplicative terms in multiple regression equations." American Journal of Political Science, 1982, 26(4), pp. 797-833.
Geda, A.; Jong, N. D. and Kimenyi, M. S. "Determinants of Poverty in Kenya: A household level Analysis," Working Paper 2005-44. Department of Economics, University of Connecticut, 2005.
Godfrey, L. G. "Instrument relevance in multivariate linear models." The Review of Economics and Statistics, 1999, 81(3), pp. 550-52.
Greer, J. and Thorbecke, E. "A Methodology for Measuring Food Poverty Applied to Kenya." Journal of Development Economics, 1986, 24, pp. 59-74.
Haiti National Strategy team. "The Export Development and Poverty Reduction Nexus: The Case of Haiti," 2006.
Handley, G.; Higgins, K.; Sharma, B.; Bird, K. and Cammack, D. "Poverty and Poverty Reduction in sub-Saharan Africa: An Overview of the Issues," Working Paper 299. London: Overseas Development Institute, 2009.
Hummels, D. and Klenow, P. L. "The Variety and Quality of a Nation's Exports." American Economic Review, 2005, 95(3), pp. 704-23.
Kenya National Bureau of Statistics. "Basic Report on Well-Being in Kenya - Based on Kenya Integrated Household and Budget Survey-2005/06," 2007.
47
Mairesse, J. and Greenan, N. "Using Employee-Level Data in a Firm-Level Econometric Study," Working Paper No. 7028. Cambridge: National Bureau of Economic Research, 1999.
Manda, D. K.; Mwabu, G. and Kimenyi, M. S. "A review of Poverty and Antipoverty Initiatives in Kenya," Working Paper No. 3. Nairobi: Kenya Institute for Public Policy Research and Analysis, 2001.
McCaig, B. "Exporting Out of Poverty: Provincial Poverty in Vietnam and U.S Market Access," Working Paper No. 502. School ofEconomics and Research School of Social Sciences, Australian National University, 2009.
McCulloch, N. and Ota, M. "Export horticulture and Poverty in Kenya," Brighton: Institute of Development Studies, University of Sussex, 2002.
Moll, P. G. "The Collapse of Primary Schooling Returns in South Africa 1960-1990." Oxford Bulletin of Economics and Statistics, 1996, 58(1), pp. 0305-9049.
Mwabu, G. "The production of Child Health in Kenya: A Structural Model of Birth Weight." Journal of African Economies, 2009, 18(2), pp. 212-60.
Mwabu, G.; Ainsworth, M. and Nyamete, A. "Quality of Medical Care and Choice of Treatment in Kenya: An Empirical Analysis." The Journal of Human Resources, 1993, 28(4), pp. 838-62.
Mwabu, G.; Kimenyi, M. S.; Kimalu, P.; Nafula, N. and Manda, D. K. "Predicting Household Poverty: A methodological Note with a Kenyan Example." Journal of African Development Review, 2003, 15(1), pp. 77-85.
Mwabu, G. and Schultz, T. P. "Wage Premiums for Education and Location of South African Workers, by Gender and Race." Economic Development and Cultural Change, 2000, 48(2), pp. 307-34.
Mwabu, G.; Wafula, M.; Gesami, R.; Kirimi, J.; Ndeng'e, G.; Kiriti, T.; Munene, F.; Chemngich, M. and Mariara, J. "Poverty in Kenya: Profile and Determinants," A Report submitted to Ministry of Finance, Kenya. Nairobi: University of Nairobi and Ministry of Finance, 2000.
Nevo, A. and Rosen, A. M. "Identification with imperfect instruments." Review of Economics and Statistics, 2010.
48
Norton, E. C.; Wang, H. and Ai, C. "Computing interaction effects and standard errors in logit and probit models." The Stata Journal, 2004, 4(2), pp. 154-67.
Oyugi, L. N. "The Determinants of Poverty in Kenya," Department of Economics. Nairobi: University of Nairobi, 2000.
Persson, T. and Tabellin, G. "Is Inequality Harmful for Growth?" American Economic Review, 1989, 84(3), pp. 600-21.
Porto, G. G. "Informal Export Barriers and Poverty " Development Research Group, The World Bank, 2004.
Quibria, M. G. and Srinivasan, T. N. "Rural poverty in Asia: priority issues and policy options," Manila: Asian Development Bank, 1991.
Ravallion, M. "Poor Areas," Washington D.C.: World Bank, 1996.
Ravallion, M. and Bidani, B. "How Robust Is a Poverty Profile?" The World Bank Economic Review, 1994, 8(1), pp. 75-102.
Reardon, T.; Delgado, C. and Matlon, P. "Determinants and effects of income diversification amongst farm households in Burkina Faso." Journal of Development Studies, 1992, 28(2), pp. 264-96.
Republic of Kenya. "The First Report on Poverty in Kenya," Ministry of Finance and Planning, 1998.
____. "Kenya Demographic and Health Survey," Nairobi, 2003.
____. "Kenya Vision 2030: The Popular Version." 2007.
____. "National Poverty Eradication Programme: 1999-2015," Government Printer, 1999.
Sen, A. "Poverty: An Ordinal Approach to Measurement." Econometrica, 1976, 44(2), pp. 219-31.
Shea, J. "Instrument relevance in multivariate linear models: A simple measure." The review of Economics and Statistics, 1997, 79(2), pp. 348-52.
Soderbom, M. and Teal, F. "Firm Size and Human Capital as Determinants of
49
Productivity and Earnings," Working Paper -2001-09. Centre for the study of African Economies, Department of Economics, University of Oxford, 2001.
____. "How can policy towards manufacturing in Africa reduce poverty? A review of the current evidence from cross-country firm studies," Centre for the study of African Economies, Department of Economics, University of Oxford, 2003.
Soderbom, M.; Teal, F.; Wambugu, A. and Kahyarara, G. "The Dynamics of Returns to Education in Kenyan and Tanzanian Manufacturing." Oxford Bulletin of Economics and Statistics, 2006, 68(3), pp. 261-88.
Staiger, D. and Stock, J. H. "Instrumental variables regression with weak instruments." Econometrica, 1997, 65(3), pp. 557-86.
StataCorp. "Stata Statistical Software: Release 10," Special Edition. College Station, Texas: StataCorp LP, 2007.
Takayama, N. "Poverty, Income Inequality, and their Measures: Professor Sen's Axiomatic Approach Reconsidered." Econometrica, 1979, 47(3), pp. 747-59.
Thon, D. "Income Inequality and Poverty: Some Problems." The review of Income and Wealth, 1981, 27(2), pp. 207-10.
____. "On Measuring Poverty." Review of income and wealth, 1979, 25(4), pp. 429-39.
United Nations Development Programme. "Macroeconomic Policy, Growth and Poverty Reduction," T. McKinley, United Nations Development Program, 2001.
Van de Walle, D. "Rural Poverty in an Emerging Market Economy: Is Diversification into Non-Farm Activities in Rural Vietnam the Solution? Public Economic Division, Policy Research Department," Washington D.C: World Bank, 1995.
Were, M. "Export-Orientation and employment patterns in Kenya's manufacturing sector," CSAE conference. 2006.
World Bank. "World Development Report, 1997: The state in a changing World," Washington D.C: World Bank, 1997.